User repository on GitHub
A polished OpenAI + TypeSafe Jev terminal interface for answers with transparent decision reports
- Forks
- 2
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
Jev System One
Python terminal app that puts Jev on both sides of an OpenAI answer. A LangGraph pipeline asks Jev for response mode, clarification and depth, has OpenAI draft, then has Jev score whether the draft answers the question and whether it carries unsupported claims before approving or sending it back.
- Stars
- 5
- Language
- Python
- License
- MIT
- Published by
- Community
- Author
- haseeb-heaven
- Last verified
- Oct 2026
About this project
A polished terminal interface where OpenAI generates useful answers and TypeSafe Jev independently evaluates their relevance, reliability, and quality. TypeSafe System One models are designed for fast, structured software decisions. Jev is the authority for response policy and quality review; it returns typed judgments, probabilities, and confidence. OpenAI never classifies the request or sets the policy—it only drafts, revises, and finalizes wording from Jev's decisions.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Jev makes typed decisions at two points in every request: policy before the draft, then a review pass that can send the draft back.
- The review is all typed calls: does the draft answer the question, are uncertain claims stated as facts, and quality on a four-level rubric.
- OpenAI never classifies the request or sets the policy, it only drafts, revises and finalizes wording from Jev's decisions.
- A full-screen Textual interface shows answer and decision panes side by side, with a one-shot ask command for scripts.
Watch out
It needs two paid providers at once, a TypeSafe key for Jev and an OpenAI key for the drafting model, and the revision loop can call both repeatedly until Jev approves or max iterations is hit, so one question is not one bill. Jev does not write the answer here, it only judges, so output quality still depends on the drafting model and its configuration. The autonomous clarification step means the human is in the loop once rather than at every ambiguity. The README reports no evaluation of whether Jev's policy calls beat simply asking the user. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
The point is separating judgment from prose. Jev picks the response mode, decides whether the question needs clarification and sets a depth policy; after OpenAI drafts, Jev judges whether the draft answers the question, whether uncertain claims are presented as facts, and how good it is. A weak draft gets sent back for revision instead of shipped.
No. If Jev flags that clarification is required, LangGraph resolves the ambiguity through OpenAI on its own and carries on. You are involved once, when you type the question, and the decisions are written to a decision trail with probabilities, confidence, model version, token usage and the TypeSafe request ID that you can read afterwards.
TypeSafe's Jev through the official Python SDK, using TypeSafeClient with Choice, Noul and Score plus a RetryPolicy, and an OpenAI model for drafting. Both keys come from a .env in the project root, alongside JEV_MODEL and LLM_MODEL, and the app adds explicit retries, timeouts and structured errors on top.
Yes. The ask command runs one question without opening the Textual interface, which is the path meant for scripts and shell workflows. The interactive side adds mouse wheel, two-finger trackpad and draggable scrollbar handling, a current-question view that replaces stale results, and an animated processing state between the Jev and OpenAI steps.
Seven. Before the draft, a Choice for response mode (factual, instructional, creative, opinion, or clarification), a Noul for clarification required, a Noul for an uncertainty notice, and a Score for response depth (brief, standard, or detailed). After the draft, a Noul for whether it answers the question, a Noul for unsupported claims, a Score for answer quality on an ordered four-level rubric, and a Noul for whether revision is required.
A maximum iteration count. LangGraph routes internally between Jev's revision verdict and OpenAI's rewrite until Jev approves or the cap is hit, and OpenAI then finalizes from the review either way. The loop is autonomous by design, which is also why one question is not one bill: both providers can be called repeatedly.
The answer and the decision trail side by side in a full-screen Textual interface. Each decision is recorded with its probability, confidence, model version, token usage, and the TypeSafe request ID, so a specific call can be traced after the fact. The report is the point of the tool as much as the answer is.
More Jev resources like this
46GitHub stars
Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.
Jev Demos & Experiments#community#jevbest#catalog-import
941GitHub stars
Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
TypeLLM/TypeLLMPython
TypeLLM/TypeLLMPython55GitHub stars
Toolkit that makes a model behind an OpenAI-compatible endpoint or running in-process answer typed choice, score, and yes/no questions from next-token probabilities in Jev API shape.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
fidecastro/jevifyPython
fidecastro/jevifyPython164GitHub stars
End-to-end toolkit that synthesizes decision data, trains and calibrates candidate scorers, evaluates fixed splits, and serves Jev-compatible inference for interactive web decisions.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
zwliJay/jev-forgePython
zwliJay/jev-forgePython70GitHub stars
Self-hosted proxy that records typed decisions, distils them into small heads on a frozen encoder, and answers locally when confident.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
bladedevoff/stuntdPython
bladedevoff/stuntdPython42GitHub stars
Open harness that renders allowed JSON answers on a diffusion canvas, denoises for a fixed step budget, then picks the top allowed tokens from final logits with no training.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
JoshuaSP/open-jevPython
JoshuaSP/open-jevPythonFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.